Problem of spurious correlation in the evaluation of steady-state carbamazepine levels using metabolite data.

The statistical problem of spurious correlation occurs when ratio variables with the same denominator are correlated. The correlation between two variables is also altered when heterogeneous groups are pooled. The problem is illustrated in the assessment of steady-state plasma levels of carbamazepine (CBZ), CBZ-10,11-epoxide (CBZ-epoxide), and 10,11-dihydro-10, 11-trans-dihydroxy-CBZ (CBZ-diol) in 12 epileptic patients on CBZ monotherapy and 12 epileptic patients on CBZ polytherapy. Expressing outcomes as dose/level ratios and metabolite level/parent drug level ratios is shown to be inappropriate, leading to spurious correlation. The amount of spurious correlation induced can be assessed by use of available statistical methodology, as is illustrated in this paper. More appropriate, alternative approaches are illustrated, using the data on the 24 patients with epilepsy.